CN-122023378-A - Unmanned aerial vehicle inspection photovoltaic panel defect detection method and system in railway line scene

CN122023378ACN 122023378 ACN122023378 ACN 122023378ACN-122023378-A

Abstract

The invention discloses a method and a system for detecting defects of an unmanned aerial vehicle inspection photovoltaic panel in a railway line scene, and belongs to the technical field of electrified railway new energy development. The method comprises the steps of carrying out image preprocessing by utilizing a retinex algorithm of self-adaptive bilateral filtering according to railway along-line photovoltaic panel images acquired by an unmanned aerial vehicle to obtain preprocessed photovoltaic panel images, and carrying out photovoltaic panel defect detection by utilizing YOLOv detection algorithm introducing a CDA-net attention mechanism according to the preprocessed photovoltaic panel images to obtain detection results. The method solves the problems of insufficient inspection and detection precision, large algorithm model and the like of the unmanned aerial vehicle, and can effectively improve the reliability and practicality of the unmanned aerial vehicle inspection on the railway along-line photovoltaic panel.

Inventors

ZHANG YUHONG
WANG ZHANGFAN
WEI YINZHONG
CHEN ZHEFENG
ZHANG SIQI
ZHOU JIANGSHAN
QIAN KANG
JIANG KE
Shi Houfu

Assignees

中国能源建设集团投资有限公司
中国能源建设集团江苏省电力设计院有限公司

Dates

Publication Date: 20260512
Application Date: 20260213

Claims (10)

1. The unmanned aerial vehicle inspection photovoltaic panel defect detection method under the scene along the railway is characterized by comprising the following steps: according to the railway along-line photovoltaic panel image acquired by the unmanned aerial vehicle, performing image preprocessing by utilizing a retinex algorithm of self-adaptive bilateral filtering to obtain a preprocessed photovoltaic panel image; and detecting defects of the photovoltaic panel by utilizing YOLOv detection algorithm introducing a CDA-net attention mechanism according to the preprocessed photovoltaic panel image to obtain a detection result.
2. The method for detecting defects of a photovoltaic panel by unmanned aerial vehicle inspection under a railway line scene according to claim 1, wherein the image preprocessing is performed by using a retinex algorithm of adaptive bilateral filtering according to a railway line photovoltaic panel image acquired by an unmanned aerial vehicle, so as to obtain a preprocessed photovoltaic panel image, and the method comprises the following steps: converting a railway along-line photovoltaic panel image acquired by an unmanned aerial vehicle from an RGB space to an HSI space to obtain H channel characteristics, S channel characteristics and I channel characteristics; carrying out self-adaptive bilateral filtering Retinex enhancement on the I channel characteristics to obtain enhanced I channel characteristics; combining the H channel characteristics, the S channel characteristics and the enhanced I channel characteristics, and converting the combined H channel characteristics, the S channel characteristics and the enhanced I channel characteristics back to an RGB space to obtain an enhanced railway along-line photovoltaic panel image; Constructing a Gaussian pyramid for the reinforced railway along-line photovoltaic panel image, carrying out self-adaptive bilateral filtering Retinex reinforcement on each layer of image of the Gaussian pyramid, and carrying out image reconstruction to obtain a reconstructed image; and carrying out illumination equalization on the reconstructed image to obtain a preprocessed photovoltaic panel image.
3. The method for detecting defects of a photovoltaic panel for inspection of a unmanned aerial vehicle in a railway line scene according to claim 2, wherein the expression of the adaptive bilateral filtering Retinex is as follows: , , , Wherein p represents a pixel point in the input image, N represents a neighborhood of the pixel point p, q represents a pixel point in the neighborhood of the pixel point p, Representing the pixel value of the pixel point p after the adaptive bilateral filtering Retinex processing, The intensity value of the pixel point p is indicated, The intensity value of the pixel point q is represented, The spatial geometry function is represented by a function of the spatial geometry, Representing the parameters of the standard deviation of the space geometry, The luminance function is represented as a function of the luminance, The standard deviation of the luminance parameter is indicated, Representing the spatial geometrical coordinates of the pixel point p, Representing the spatial geometry of the pixel point q.
4. The method for detecting defects of unmanned aerial vehicle inspection photovoltaic panels in a railway line scene as claimed in claim 3, wherein the brightness standard deviation parameter is Calculated by the following formula: , Wherein, the Coefficients representing the constrained luminance standard deviation parameters, Representing the intensity of the edge(s), , Representing the sum of the edge pixels of the input image, , Representing the image after edge detection by the canny algorithm, Representing the number of pixels of the input image.
5. The method for detecting defects of unmanned aerial vehicle inspection photovoltaic panels in a railway line scene as claimed in claim 3, wherein the space geometric standard deviation parameter is as follows The method comprises the following steps: Calculating noise variance of input image Repeating the following steps until the space geometric standard deviation parameter is output : According to the variance of noise Calculating a eigenvalue threshold using a gamma cumulative distribution function ; Traversing the input image by using a preset sliding window of S multiplied by S, calculating a gradient covariance matrix of each window area, and if the maximum eigenvalue of the gradient covariance matrix is smaller than or equal to Judging the window area as a weak texture area to obtain a weak texture area set w; calculating noise variance of weak texture region set w If (if) Order in principle Outputting the geometric standard deviation parameter of the space Otherwise, let Wherein, the method comprises the steps of, For a preset tolerance, a is a coefficient of constraint space geometric standard deviation parameter.
6. The method for detecting defects of unmanned aerial vehicle inspection photovoltaic panels in a railway line scene as claimed in claim 5, wherein the characteristic value threshold value is Calculated by the following formula: , , , Wherein, the An inverse function representing the gamma cumulative distribution function, Representing the maximum cumulative distribution probability that the distribution is at the maximum, And Is a parameter of the gamma distribution and, The number of pixels of the input image is represented, , Representing the covariance matrix of the input image, Representation of Is used to determine the minimum characteristic value of the set of values, Operator matrix for derivative of x-axis direction Is used for the track of (a), Representing the derivative of the input image in the x-axis direction; the gradient covariance matrix is calculated by the following formula: , , Wherein, the Representing a gradient matrix Is used for the co-variance matrix of (a), In the event of a window area, Representing window regions The derivative in the x-axis direction, Representing window regions The derivative in the direction of the y-axis, Representing a gradient covariance matrix Is used for the maximum characteristic value of the (c), Representing a gradient covariance matrix Is used to determine the minimum characteristic value of the (c), Representing window regions Is described.
7. The method for detecting defects of a photovoltaic panel for unmanned aerial vehicle inspection in a scene along a railway according to claim 2, wherein the performing adaptive bilateral filtering Retinex enhancement on the I-channel characteristics to obtain enhanced I-channel characteristics comprises: estimating an illumination component of the I channel feature by utilizing an adaptive bilateral filtering Retinex, and separating a reflection component from the I channel feature according to the estimated illumination component; nonlinear stretching is carried out on the reflection component by using a sigmoid function, and the reflection component after nonlinear stretching is obtained; Combining the nonlinear stretched reflection component and the estimated illuminance component to obtain enhanced I channel characteristics; the nonlinear stretching of the reflected component using a sigmoid function is achieved by the following formula: , Wherein, the Representing nonlinear pre-stretch coordinates The reflected component at which the reflection is generated, Representing nonlinear post-stretching coordinates The reflected component at which the reflection is generated, In order to set the value of the preset value, Representing a nonlinear tensile strength parameter.
8. The method for detecting defects of unmanned aerial vehicle inspection photovoltaic panels in a railway line scene according to claim 2, wherein the illumination balancing is realized by the following formula: , Wherein, the Representing the illumination-balanced front coordinates The component of the illumination at which, For the coordinates after illumination balance The component of the illumination at which, And To adjust the coefficients.
9. The method for detecting defects of a photovoltaic panel for unmanned aerial vehicle inspection in a scene along a railway line according to claim 1, wherein the YOLOv detection algorithm introducing a CDA-net attention mechanism introduces the CDA-net attention mechanism between a Neck module and a detection head module of a YOLOv8 model; the CDA-net attention mechanism includes a spatial attention module and a channel attention module.
10. Unmanned aerial vehicle inspection photovoltaic board defect detection system under railway line scene, its characterized in that includes: The image preprocessing module is used for preprocessing the image by utilizing a retinex algorithm of self-adaptive bilateral filtering according to the railway along-line photovoltaic panel image acquired by the unmanned aerial vehicle to obtain a preprocessed photovoltaic panel image; and the defect detection module is used for detecting the defects of the photovoltaic panel by utilizing YOLOv detection algorithm which introduces a CDA-net attention mechanism according to the preprocessed photovoltaic panel image to obtain a detection result.

Description

Unmanned aerial vehicle inspection photovoltaic panel defect detection method and system in railway line scene Technical Field The invention relates to the technical field of new energy development of electrified railways, in particular to a method and a system for detecting defects of an unmanned aerial vehicle inspection photovoltaic panel in a scene along a railway. Background The rapid development of electrified railways brings convenience to the people to travel and simultaneously leads to the rapid increase of the energy consumption requirements. As far as the full life cycle energy usage pattern of the current electrified railway is concerned, it is still one of the important industries of carbon emission. The photovoltaic power generation system can be built by utilizing the side slope along the railway, so that the carbon emission level of the railway industry can be effectively reduced. Because the length of the railway line is longer, the railway line covers various topography and landforms, and the manual inspection difficulty is higher. The inspection unmanned aerial vehicle carrying the simple cleaning tool is difficult to automatically distinguish the difference between the falling She Huichen and the crack hot spots, so that the operation and maintenance difficulty of the railway along-line photovoltaic panel is high, and the new energy development in the railway field is restricted. At present, the application of the neural network in defect detection is mature, but the main stream algorithm with high detection precision occupies more memory at present. Due to the limitation of the capacity of a built-in chip of the unmanned aerial vehicle and the endurance of a battery, the main stream algorithm cannot be adapted. Disclosure of Invention The invention aims to overcome the problems in the prior art, and provides a method and a system for detecting defects of a photovoltaic panel patrolled and examined by an unmanned aerial vehicle in a scene along a railway, which are used for preprocessing railway along-line photovoltaic panel images acquired by the unmanned aerial vehicle through a retinex algorithm of self-adaptive bilateral filtering, and then detecting the problems existing in the photovoltaic panel through a YOLOv detection algorithm introducing a CDA-net attention mechanism and giving an overhaul instruction. In order to solve the technical problems, the invention is realized by adopting the following technical scheme: In a first aspect, the invention provides a method for detecting defects of a photovoltaic panel by unmanned aerial vehicle inspection under a scene along a railway, comprising the following steps: according to the railway along-line photovoltaic panel image acquired by the unmanned aerial vehicle, performing image preprocessing by utilizing a retinex algorithm of self-adaptive bilateral filtering to obtain a preprocessed photovoltaic panel image; and detecting defects of the photovoltaic panel by utilizing YOLOv detection algorithm introducing a CDA-net attention mechanism according to the preprocessed photovoltaic panel image to obtain a detection result. Optionally, the preprocessing of the image is performed by using a retinex algorithm of adaptive bilateral filtering according to the railway along-line photovoltaic panel image acquired by the unmanned aerial vehicle, so as to obtain a preprocessed photovoltaic panel image, including: converting a railway along-line photovoltaic panel image acquired by an unmanned aerial vehicle from an RGB space to an HSI space to obtain H channel characteristics, S channel characteristics and I channel characteristics; carrying out self-adaptive bilateral filtering Retinex enhancement on the I channel characteristics to obtain enhanced I channel characteristics; combining the H channel characteristics, the S channel characteristics and the enhanced I channel characteristics, and converting the combined H channel characteristics, the S channel characteristics and the enhanced I channel characteristics back to an RGB space to obtain an enhanced railway along-line photovoltaic panel image; Constructing a Gaussian pyramid for the reinforced railway along-line photovoltaic panel image, carrying out self-adaptive bilateral filtering Retinex reinforcement on each layer of image of the Gaussian pyramid, and carrying out image reconstruction to obtain a reconstructed image; and carrying out illumination equalization on the reconstructed image to obtain a preprocessed photovoltaic panel image. Optionally, the expression of the adaptive bilateral filtering Retinex is as follows: , , , Wherein p represents a pixel point in the input image, N represents a neighborhood of the pixel point p, q represents a pixel point in the neighborhood of the pixel point p, Representing the pixel value of the pixel point p after the adaptive bilateral filtering Retinex processing,The intensity value of the pixel point p is indicated,The intensity value of the